Asynchronous transfer mode (ATM) is gaining rapid recognition as the technology of choice for the transmission of broadband information. The fixed length cell size employed in ATM technology supports the delivery of a wide range of multimedia information including speech, video and data. In order to service this wide variety of information the ATM Traffic Management Forum has specified that ATM traffic be divided into five service categories. Two of these service categories are for real-time transmission, namely constant bit rate (CBR) and real-time variable bit rate (rtVBR). There are, as well, three non-real-time categories, namely non-real-time variable bit rate (nrt-VBR); unspecified bit rate (UBR) and available bit rate (ABR)
As the various service categories are designed to carry different types of information, different Quality of Service (QoS) parameters apply. The ATM Traffic Management Forum has listed six Quality of Service parameters namely: peak-to-peak Cell Delay Variation (peak-to-peak CDV); maximum Cell Transfer Delay (Max CTD); Cell Loss Ratio (CLR); Cell Error Ratio (CER); Severely Errored Cell Block Ratio (SECBR) and Cell Misinsertion Ratio (CMR). Of these six, two are of greatest concern when scheduling. These are the maximum Cell Transfer Delay (Max CTD) and Cell Delay Variation (CDV). Traffic of different service classes, and even traffic of the same service class may have different delay requirements. Fair scheduling between connections of the various ATM service categories onto a single resource while meeting the multiple Quality of Service delay guarantees is difficult to achieve.
One prior art method for scheduling such traffic, known as shaping, involves calculating a theoretical emission time (TET) for each active connection and attempting to service each connection as close to that theoretical emission time as possible. Unfortunately, most connections cannot be serviced exactly at their TET due to finite output time requirements for the shared resource, as well as scheduling collisions where multiple connections are scheduled to output at the same time. Though the finite output times cannot be avoided, the effects of scheduling collisions can and should be reduced through prioritization. Without prioritization, a connection with loose delay requirements may actually get serviced prior to a connection with tight delay requirements scheduled at the same TET.
One proposed solution to this problem is to use two or more shaping devices (or shapers) to schedule the connections where connections with similar Quality of Service delay requirements are scheduled in the same shaper. Exhaustive servicing is commonly used to determine which shaper gets to send a cell at any given time. Prioritized shaping is discussed in U.S. application Ser. No. 08/873,064 which is based on provisional application 60/020,642 filed Jun. 27, 1996. The contents of the aforementioned application are incorporated herein by reference. When more than one calendar (as defined in the aforementioned application) contains a connection ready to send a cell, only the highest priority calendar gets to transmit. All lower priority calendars have to wait until all calendars with higher priority have nothing ready to send. This calendar prioritization can be based upon the CDV and possibly max CTD requirements of the traffic scheduled in each calendar. There are, however, certain limitations which can be identified in connection with the aforementioned multiple shaper or calendar solution. For example, there is no quantitative link between the CDV and max CTD requirements between each shaper's traffic and the prioritization of that shaper respective of all other shapers. When a high priority shaper schedules several connections over a relatively short transient period, lower priority shapers have to wait until this transient finishes before they can resume sending cells from their connections. During these transients, low priority connections may violate their delay requirements while those scheduled on high priority shapers may actually beat their delay requirements. Thus, high priority connections may actually better their QoS at the expense of lower priority connections. To properly support QoS delay requirements, shaper priorities must be dynamically assigned based upon the amount of time that their connections have been delayed from ideal theoretical emission times.
Further, heretofore traffic is either shaped or placed in work-conserving, weighted-fair-queues. Shaping works well for real-time traffic where rates are explicitly guaranteed and weighted-fair-queuing works well for non-real-time traffic that does not need to be prioritized or rate limited. However, most non-real-time traffic has an associated minimum rate that must be guaranteed and a maximum rate that cannot be exceeded. It is also often desirable to provide work conservation between these two limits to ensure full resource usage as precise instantaneous bandwidth availability is rarely known for non-real-time traffic. An example would be non-real-time variable bit rate where the minimum guaranteed scheduled rate is sustainable cell rate (SCR) though output rate will be lower when an upstream source sends below SCR. The maximum rate is peak cell rate (PCR) with an associated burst tolerance (BT) that must be respected. In this case it is not desirable to simply shape up to PCR as this may infringe upon rate guarantees of the available bit rate (ABR) and possibly unspecified bit rate (UBR) connections. On the other hand, it is also not desirable to simply shape to SCR as this does not allow the non-real-time variable bit rate connection to make use of any additional unused bandwidth. This minimum rate guarantee with work conservation up to a maximum rate cannot be provided for by shaping or weighted-fair-queuing alone.
Furthermore, scheduling using a single device, namely a shaper or a weighted-fair-queue, does not provide proper support for rate-based backpressure where all non-real-time connections are throttled back to minimum required rates. Weighted-fair-queuing by itself is not able to throttle back all connections due to its work-conserving nature, while shaping is far too slow to respond. For a shaper to throttle back the rates of all connections scheduled, these connections would either have to be descheduled and rescheduled with a longer period or the shaper would have to wait for each connection to be serviced before rescheduling it with a longer period. At the conclusion of the backpressure, the same procedure would have to be followed for increasing the rates of the connections. As thousands of connections may be scheduled these procedures would take a very long time to complete. The net effect is an increased delay in the control loop which is therefore harder to stabilize.